Workpiece Recognition And Location Based On Point Cloud Processing Technology

In the field of industrial production and transportation,intelligent robots are used for sorting and producing instead of human beings,which has been a subject of great concern in recent years.In this subject,it is a key link to identify and locate industrial components on the production line.This thesis is aimed at the detection and location of industrial components in industrial production.In addition,the point cloud data processing technology and position estimation are studied.The main contents are as follows.(1)A scheme based on 3D object recognition and robot grabbing is designed.Offline,it collects point cloud of the industrial components to be identified.Through denoising,splicing and the feature edge enhancing of point cloud,the point cloud of the industrial component is made into model and stored in model database.Online,it collects the point cloud data on the pipeline in real time and uses target recognition algorithm to identify and locate the target.The system’s offline and online mode is suitable for the identification of other objects,with a certain degree of versatility.(2)It designs a target recongnition algorithm based on local feature descriptor and hough voting.The key points are extracted through the key point extraction based on ISS3 D and then described by the local feature descriptor Ro PS.The hypothesis of the target is generated by Hough voting,and the hypothesis instance is determined as part of the target after the improved ICP algorithm.The experimental results show that the system improves the recognition efficiency by the optimization of feature matching,segmentation of target region and so on,besides satisfying the accuracy and robustness.(3)The object recognition technology of point cloud is studied.The principal component analysis method based on Gauss weight is used to solve the normal vector of point cloud to fix the sparseness,and the edge features of point cloud are emphasized by Weight LOP algorithm in the end.Random sampling consistency algorithm combined with feature distance is used to purify point pairs to tackle the problem of matching wrong features.To solve the problem of incompleteness,the double RANSAC algorithm is used to optimize the feature matching pairs in the hypothesis generation method,and an adaptive fine-tuning method is designed for the two parameters of Hough space.The experimental results show that the proposed hypothesis generation method and the double hypothesis verification can effectively eliminate the false points and obtain the recognition target.(4)It estimates the position of the target by fitting the edge line based on random sampling consensus algorithm.